{"id":129071,"date":"2026-08-04T13:18:08","date_gmt":"2026-08-04T13:18:08","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/129071\/"},"modified":"2026-08-04T13:18:08","modified_gmt":"2026-08-04T13:18:08","slug":"antitrust-uncertainty-and-ai-security-collaboration","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/129071\/","title":{"rendered":"Antitrust Uncertainty and AI Security Collaboration"},"content":{"rendered":"

The\u00a0clock is ticking as the U.S. government scrambles\u00a0to\u00a0prevent\u00a0the cybersecurity capabilities\u00a0of frontier AI models\u00a0from\u00a0being misused against\u00a0government operations, critical infrastructure, and U.S. private sector\u00a0networks.\u00a0With Chinese labs only\u00a0months<\/a>\u00a0behind U.S.\u00a0companies,\u00a0advanced\u00a0cyber capabilities\u00a0similar to\u00a0those\u00a0displayed by\u00a0Mythos and GPT-5.6\u00a0may\u00a0soon<\/a>\u00a0be available to\u00a0criminals and intelligence services around the world.\u00a0Distillation of U.S. models\u2019 capabilities\u00a0is\u00a0helping<\/a>\u00a0Chinese labs\u00a0remain\u00a0\u201cfast followers\u201d and reducing the\u00a0U.S.\u00a0lead.\u00a0Yet\u00a0U.S.\u00a0companies are not collaborating with each other to stop this\u00a0adversarial\u00a0extraction\u00a0and replication\u00a0of technical advances, even though a joint effort is\u00a0required. The reason: fears that\u00a0some of the necessary collaboration could violate\u00a0U.S.\u00a0antitrust law.\u00a0\u00a0<\/p>\n

Collaboration on risks posed by AI models themselves has also been limited by concerns about antitrust law. AI models are increasingly demonstrating capabilities that could be misused in developing <\/a>biological weapons or exploiting<\/a> cybersecurity vulnerabilities. Even maintaining oversight and control of increasingly capable AI models is becoming more challenging<\/a>. Although competitive pressures and personal rivalries may also limit the extent to which the leading companies are willing to work together, antitrust fears are currently a central factor deterring<\/a> AI developers from collaborating to reduce these risks or to ensure they take enough time to vet new models.\u00a0 \u00a0 <\/p>\n

These\u00a0concerns\u00a0represent\u00a0a\u00a0collective-action\u00a0problem.\u00a0Pressure to release more capable models reduces the time available\u00a0for a company\u00a0to build in\u00a0safeguards and evaluate risks, lest it lose ground to competitors, and an inability\u00a0to take joint action against distillation\u00a0facilitates\u00a0accelerated diffusion of advanced capabilities to\u00a0adversaries.\u00a0The same remedy would help\u00a0ameliorate\u00a0both\u00a0aspects of the\u00a0problem:\u00a0providing\u00a0a firm legal basis for collaboration.\u00a0<\/p>\n

A\u00a0bipartisan bill<\/a>\u00a0has now been\u00a0proposed<\/a>\u00a0that would provide targeted protections to collaboration on AI security risks, including\u00a0adversarial distillation.\u00a0The Collaboration on Adversarial Threats and Security Risks Act\u00a0builds on the precedent set in 2015 for collaboration on cybersecurity issues.\u00a0It\u00a0would\u00a0provide clarity to ensure that\u00a0antitrust\u00a0law\u00a0does not deter\u00a0legitimate collaboration\u00a0while\u00a0retaining\u00a0safeguards against anticompetitive behavior.\u00a0<\/p>\n

Antitrust Uncertainty Deters Beneficial Collaboration\u00a0<\/p>\n

Because AI is developing so rapidly, collaboration on security risks needs to be possible on short timelines and in a flexible manner. Technical researchers may identify risks that also affect other companies\u2019 models and that need to be addressed before hostile actors can take advantage of them. Yet analysis of antitrust risks can be highly fact-specific, and time-intensive review by counsel can deter busy researchers from seeking authorization to collaborate. The law is not clear enough for in-house lawyers to authorize technical researchers to reach out directly to counterparts at other companies to collaborate on security issues. The barrier may be even more daunting for smaller startups that have fewer resources to engage in legal reviews that distract from their core work. \u00a0<\/p>\n

For years, U.S. frontier labs such as\u00a0OpenAI<\/a>,\u00a0Google DeepMind<\/a>, and\u00a0Anthropic<\/a>\u00a0have\u00a0called for government action,\u00a0suggesting\u00a0that\u00a0antitrust concerns serve as a barrier to collaboration on security-related risks.\u00a0Similarly, in off-the-record fora, employees of some leading developers\u00a0frequently\u00a0cite\u00a0in-house counsels\u2019\u00a0concerns about antitrust risks\u00a0as a\u00a0central reason that more collaboration does not occur.\u00a0A\u00a02024 analysis<\/a>\u00a0by the Institute for Law & AI argued that\u00a0\u201c[i]n the absence of some sort of guidance or safe harbor, the risk-averse in-house legal teams at leading AI companies \u2026 are unlikely to allow any significant cooperation or communication between\u00a0rank and file\u00a0employees.\u201d\u00a0<\/p>\n

The problem was compounded in late 2024, when the\u00a0Department of Justice and the Federal Trade Commission\u00a0(FTC)\u00a0withdrew<\/a>\u00a0the\u00a02000\u00a0Antitrust Guidelines for Collaborations Among Competitors<\/a>. Although the Guidelines were generic and not specific to the AI context, they provided some additional reassurance regarding how collaborations are viewed from the enforcement perspective. Their absence now creates even more uncertainty.\u00a0<\/p>\n

Tools used in other areas to reduce antitrust risks are often not helpful in the AI context because of the fast-moving nature of the research. Formalized contractual arrangements will often not be helpful or appropriate, so the notification process<\/a> under the National Cooperative Research and Production Act will generally not be available. Similarly, the procedures for seeking FTC advisory opinions<\/a> or DOJ business review determinations<\/a> often take months\u2014too slow to provide useful guidance when collaboration is urgent.\u00a0 <\/p>\n

Some limited collaboration among frontier labs does still\u00a0occur. Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI\u00a0are all members of the\u00a0Frontier Model Forum<\/a>, which\u00a0has\u00a0facilitated\u00a0a\u00a0series<\/a>\u00a0of information-sharing efforts among its members.\u00a0But FMF\u2019s membership is limited.\u00a0SpaceXAI\u00a0is not a member, and no startups or smaller companies are members\u00a0either.\u00a0<\/p>\n

Adversarial Distillation of U.S. AI Models\u00a0<\/p>\n

The lack of a coordinated response to adversarial distillation campaigns provides a useful case study on how antitrust fears deter beneficial collaboration.\u00a0Earlier this year,\u00a0Google<\/a>,\u00a0OpenAI<\/a>, and\u00a0Anthropic<\/a> each disclosed information that publicly revealed a pattern of Chinese industrial-scale distillation attacks. These U.S. companies\u2019 most advanced models are \u201cclosed-weight\u201d models that are accessed on the web, via an app, or through an application programming interface (API) that lets other programs send requests to the model. \u00a0<\/p>\n

Chinese labs have been using deceptive techniques to access these U.S. models and elicit outputs that they use to train new models with similar capabilities,\u00a0then\u00a0release\u00a0them\u00a0as \u201copen-weight\u201d models. Because open-weight models\u00a0can be freely downloaded, altered, and used locally,\u00a0they pose\u00a0a wider range of risks. Providers of closed models\u00a0can use tools\u00a0such as\u00a0classifiers<\/a>\u00a0or\u00a0monitors<\/a>\u00a0to\u00a0identify\u00a0and block attempts to misuse the model, but\u00a0these types of external\u00a0safeguards\u00a0do not constrain open-weight models.\u00a0Similarly,\u00a0the\u00a0internal\u00a0safeguards that may be built into the closed-weight models\u00a0do not reliably<\/a>\u00a0accompany those\u00a0models\u2019\u00a0capabilities\u00a0into\u00a0distilled models.\u00a0Moreover,\u00a0even\u00a0to the extent that\u00a0an\u00a0open-weight\u00a0model\u00a0is\u00a0trained to refuse certain queries, that training can be\u00a0cheaply fine-tuned<\/a>\u00a0away.\u00a0<\/p>\n

The\u00a0executive\u00a0director of the FMF, Chris Meserole, recently\u00a0testified<\/a> that U.S. models once had a 12-to 18-month lead over foreign models, but that distillation attacks have helped narrow that lead to 4 to 6 months before similar capabilities emerge. The United Kingdom\u2019s AI Security Institute has similarly identified<\/a>\u00a0a narrowing of\u00a0the gap between closed-weight and open-weight models\u00a0on cyber capabilities in particular\u2014from 6\u00a0to\u00a010 months in 2025 to 4\u00a0to\u00a07 months\u00a0in mid-2026.\u00a0<\/p>\n

When Anthropic released a preview version of its Mythos model, which represented a significant leap<\/a> in cyber capabilities, it gave limited access<\/a> to a trusted set of partners who used the model for bolstering defenses. (Similarly, OpenAI initially released<\/a> GPT-5.6 to a \u201csmall group of trusted partners\u201d at the U.S. government\u2019s request.) A broader release of Mythos would not only have increased the risk of direct misuse but would have posed the risk of the capabilities being distilled. Over time, the executive branch\u2019s concerns about the capabilities of Mythos and the company\u2019s more-constrained Fable model led to the imposition<\/a> of export controls on the models for 18 days<\/a>. But despite precautions, Michael Kratsios, director of the Office of Science and Technology Policy at the White House, has said<\/a> that the Chinese company Moonshot AI distilled Fable in developing its Kimi K3 model.<\/p>\n

In his testimony, the FMF\u2019s Meserole said that\u00a0the challenge in countering distillation attacks is that the necessary information \u201cis distributed amongst a wide array of industry actors.\u201d\u00a0The Center for a New American Security\u00a0(CNAS)\u00a0has\u00a0identified<\/a>\u00a0information such as \u201caccount indicators, network origins, behavioral patterns, and hashed prompts\u201d that could usefully be shared among\u00a0U.S. companies trying to combat distillation.\u00a0Yet, as Meserole stated,\u00a0FMF members\u00a0have \u201chad to take a fairly conservative approach under antitrust law to even have a conversation about how to\u00a0identify\u00a0distillation\u201d\u00a0and\u00a0\u201chave not had a conversation about how to\u00a0counter\u00a0it, given existing antitrust concerns.\u201d\u00a0<\/p>\n

The Need\u00a0to Enable\u00a0Coordination\u00a0<\/p>\n

A robust effort to\u00a0combat\u00a0distillation\u2014and thereby extend the period\u00a0before open-weight models make advanced cyber capabilities widely available\u2014may require collaboration beyond information sharing.\u00a0A forthcoming report from\u00a0the Institute for Progress\u2019s\u00a0Saif Khan<\/a> will analyze whether the new precedent for limited rollouts of cyber-capable models (like Mythos and GPT-5.6) could also help partially rebuild the lead that the U.S. models have over Chinese models. Even if adversarial distillation is unavoidable, limiting access to only trusted partners for an initial period could delay it (and thus delay China\u2019s access to frontier-level capabilities). But if multiple U.S. companies are developing comparable new models, access to any one of them could be enough for Chinese labs to build their own via distillation. Thus, coordination would be necessary for such an approach to be effective\u2014but such coordination could resemble an output restraint, potentially a per se violation of antitrust law.\u00a0<\/p>\n

The need\u00a0to\u00a0enable coordination\u00a0does not rest solely on the threat from China.\u00a0Antitrust law could also prevent U.S. companies from coordinating to provide the U.S. government with greater access to unreleased models. Although\u00a0a\u00a0recent\u00a0executive order<\/a>\u00a0called for a \u201cvoluntary\u201d 30-day review by the government prior to the release of new frontier models, earlier reports\u00a0suggested<\/a>\u00a0that a 90-day window was considered. If several companies wanted to provide more than\u00a030 days\u2019 access but were worried that doing so unilaterally would disadvantage them, coordination could help the government get greater access. Similarly, given the recent\u00a0incident<\/a>\u00a0in which an OpenAI model hacked into Hugging Face\u2019s servers during evaluations, companies could\u00a0reasonably\u00a0determine\u00a0that U.S. government evaluation of models at earlier stages\u2014rather than only\u00a0immediately\u00a0before release\u2014would be beneficial. In either situation, though, antitrust fears would\u00a0likely prevent\u00a0companies from\u00a0coordinating on\u00a0such an approach, despite the possible benefits to the public.\u00a0Even under a \u201crule of reason\u201d analysis\u2014a fact-specific inquiry into the anticompetitive effects and procompetitive justifications for a restraint on competition\u2014public safety justifications for\u00a0such\u00a0restraints\u00a0do not fare well<\/a>.\u00a0Thus,\u00a0congressional action is needed to ensure that beneficial collaboration does not fall afoul of\u00a0existing antitrust doctrine.\u00a0<\/p>\n

Following the Cybersecurity Precedent\u00a0<\/p>\n

Just as is the case now with AI security risks,\u00a0antitrust fears once\u00a0created\u00a0private sector reluctance to\u00a0collaborate on cybersecurity problems.\u00a0In 2014,\u00a0DOJ\u00a0and the\u00a0FTC\u00a0found<\/a>\u00a0that\u00a0\u201c[s]ome\u00a0private entities [were] hesitant to share cyber threat information with each other, especially competitors, because they have been counseled that sharing of information among competitors may raise\u00a0antitrust concerns.\u201d\u00a0To counter this fear, the\u00a0agencies issued a\u00a0joint policy statement,\u00a0unambiguously proclaiming\u00a0that\u00a0they\u00a0\u201cdo not believe that antitrust is \u2013 or should be \u2013 a roadblock to legitimate cybersecurity information sharing.\u201d\u00a0\u00a0<\/p>\n

A similar approach by the agencies would be a good first step here. As with cybersecurity information sharing, some amount of information sharing regarding AI security risks is presumably permissible under existing law, but uncertainty regarding enforcement risk is a deterrent. In April, the Trump administration pledged<\/a>\u00a0to\u00a0\u201c[e]nable\u00a0the private sector to better coordinate against [distillation] attacks.\u201d\u00a0Such guidance, addressing not only distillation but other AI-related security risks,\u00a0could be provided\u00a0in the context of replacing\u00a0the 2000 Guidelines\u2014a project on which\u00a0DOJ and\u00a0the\u00a0FTC\u00a0sought public comments<\/a>\u2014or as a standalone, sector-specific guidance document.\u00a0A number of\u00a0comment letters urged the agencies to provide guidance in this area, including\u00a0those from\u00a0RAND<\/a>,\u00a0the\u00a0Council on Strategic Risks<\/a>,\u00a0AEI\u2019s\u00a0Will Rinehart<\/a>,\u00a0the Mercatus Center\u2019s\u00a0Alden Abbott<\/a>,\u00a0the\u00a0Competitive Enterprise Institute<\/a>,\u00a0and the\u00a0Law Reform Institute<\/a>.\u00a0CNAS has also\u00a0advocated<\/a>\u00a0for such guidance.\u00a0<\/p>\n

But agency guidance\u2014though valuable\u2014would not suffice. A statement on the enforcement perspectives of federal agencies would not protect companies from private plaintiffs or claims under state law. Even after the 2014 cybersecurity policy statement, sufficient uncertainty remained that legislation was needed. The following year, the Cybersecurity Information Sharing Act of 2015 codified and expanded the agencies\u2019 view, providing<\/a> that \u201cit shall not be considered a violation of any provision of antitrust laws\u201d for private entities to share certain cybersecurity-related information\u2014or to provide related assistance to each other\u2014when done for cybersecurity purposes. These statutory protections may not have significantly changed the boundaries of what antitrust law permitted, but they dramatically improved clarity and certainty. Yet CISA, focused on traditional cybersecurity issues, does not cover the current set of risks.\u00a0<\/p>\n

A Legislative Solution\u00a0<\/p>\n

To facilitate needed private-sector collaboration on AI security risks, legislation is urgently needed to clarify what types of collaboration are permissible. To that end, the Collaboration on Adversarial Threats and Security Risks Act<\/a> was introduced on July 23 by Senators Adam Schiff (D-CA) and Jim Banks (R-IN) and Representatives Bob Latta (R-OH) and George Whitesides (D-CA).\u00a0\u00a0<\/p>\n

The Act would take a similar approach to the protections granted in CISA.\u00a0The core of the Act is a list of\u00a0six categories of \u201ccovered artificial intelligence security risks\u201d\u00a0that may occur during the \u201cdevelopment, training, testing, evaluation, deployment, use, or release\u201d of AI:\u00a0<\/p>\n

Weaponization or theft,\u00a0including distillation,\u00a0by\u00a0a\u00a0covered nation (China, Russia, North Korea, or Iran)\u00a0in a manner that poses a significant threat to national security;\u00a0<\/p>\n

Facilitation of a\u00a0chemical, biological, radiological, nuclear, or offensive cyber weapon;\u00a0<\/p>\n

Certain serious risks to critical infrastructure;\u00a0<\/p>\n

Substantial reductions in the ability to oversee or disable AI;\u00a0<\/p>\n

Autonomous improvement of AI in a manner that poses one of the\u00a0above\u00a0risks; and\u00a0<\/p>\n

Vulnerability to unauthorized access that poses one of the\u00a0above\u00a0risks or is\u00a0directed by a covered nation.\u00a0<\/p>\n

Analogous to CISA, the Act would protect information-sharing and\u00a0assistance\u00a0aimed at addressing those risks. It would also\u00a0protect\u00a0efforts to\u00a0address those risks by\u00a0coordinating\u00a0on\u00a0delays in\u00a0developing or releasing AI. The Act would thus cover the\u00a0types of scenarios described above, in which companies may want to coordinate to give the U.S. government earlier access to models or to collaborate on combatting adversarial distillation by\u00a0limiting the\u00a0initial\u00a0release of models.\u00a0These protections would be available\u00a0not just for large companies like FMF members but also for smaller developers whose current routes for collaboration on security risks are even more limited.\u00a0<\/p>\n

The Act also contains safeguards to ensure the antitrust protections are not abused. Companies would have the burden of proving, as an affirmative defense, that they acted in good faith and for the exclusive purpose of addressing covered risks. They would have to implement reasonable internal controls to limit the extent to which any information or assistance they receive can be used for other purposes. For coordinated delays in particular, companies would have to notify DOJ before taking action. Additionally, DOJ could seek injunctive relief when a company\u2019s action violates antitrust laws\u2014and even if a company successfully demonstrates that it acted in good faith to address a covered risk, its conduct could still be enjoined if it is reasonably likely to increase\u00a0those risks instead.\u00a0\u00a0<\/p>\n

Taken together, this set of safeguards would make it highly unlikely that the Act\u2019s protections could be abused for anticompetitive ends.\u00a0Coordination\u00a0beyond\u00a0good-faith efforts to address covered risks\u00a0would fall entirely outside the Act\u2019s protections and would be subject to all existing remedies.\u00a0<\/p>\n

Moving\u00a0Forward, Urgently\u00a0<\/p>\n

Although the \u201cMythos moment\u201d was triggered by cybersecurity capabilities,\u00a0sequels will\u00a0likely arrive\u00a0soon enough in other areas\u00a0such as biological capabilities.<\/a>\u00a0Thus,\u00a0quick action is needed\u00a0to enable collaboration\u00a0between American companies.\u00a0DOJ and the FTC\u00a0issued<\/a>\u00a0their cybersecurity guidance\u00a0in April 2014, and CISA\u00a0was\u00a0signed into law<\/a>\u00a0in December 2015, only\u00a020\u00a0months later. But\u00a0the agencies have not yet issued any guidance\u00a0for collaboration on AI security risks, and\u00a0even if they did so tomorrow, the\u00a0pace<\/a> of AI development means that the next 20 months will probably be full of challenges that individual companies cannot address adequately without a targeted legislative framework that facilitates coordination. By acting quickly, Congress can significantly improve the odds that U.S. companies rise to the occasion in future high-risk Mythos moments.\u00a0<\/p>\n

Authors\u2019\u00a0Note: Through the Law Reform Institute,\u00a0the authors\u00a0developed a legislative\u00a0proposal<\/a>\u00a0on this topic\u00a0last year.\u00a0LRI\u00a0provided input on the Collaboration on Adversarial Threats and Security Risks Act to the sponsors\u2019 offices\u00a0and has expressed support for the bill.\u00a0<\/p>\n

FEATURED IMAGE: The U.S. Capitol building is seen on July 25, 2026 in Washington, DC. (Photo by Kevin Carter\/Getty Images)<\/p>\n","protected":false},"excerpt":{"rendered":"The\u00a0clock is ticking as the U.S. government scrambles\u00a0to\u00a0prevent\u00a0the cybersecurity capabilities\u00a0of frontier AI models\u00a0from\u00a0being misused against\u00a0government operations, critical infrastructure,…\n","protected":false},"author":2,"featured_media":129072,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[24,5103,25,111,1069,1670,12722,313,1715],"class_list":["post-129071","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai","tag-ai","tag-antitrust","tag-artificial-intelligence","tag-artificial-intelligence-ai","tag-big-tech","tag-congress","tag-cyberattacks","tag-cybersecurity","tag-regulation"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/129071","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=129071"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/129071\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/129072"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=129071"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=129071"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=129071"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}